I started using ChatGPT Plus1 seven months ago, and I'm surprised at how much I rely on it for day-to-day tasks. I don't need to Google as much as before, as answers tend to solve my immediate needs.
Here are a few use cases I found extremely useful.
I find it mesmerizing to generate visuals and images with it — I have several parallel side projects on this, which I haven't published yet. But that's a story for another day.
Ethical and safety concerns aside, I find it quite promising that, relatively soon, we'll have models with similar performance to GPT-4 and DALL-E 3 running locally on our devices without a subscription.
I haven't measured how much more efficient as a software developer it's made me. Yet after hundreds of conversations ChatGPT has saved me some "human compute hours" and helped me develop several code solutions faster. In this case, twenty dollars a month is a no-brainer.
"We shape our tools and thereafter our tools shape us." Said John Culkin2. Language models are rapidly changing how we interact with machines. The next step could be removing humans from the loop and delegating digital tasks to machines.
Pulling a mini-essay and sketch weekly is not an easy feat.
I've been doing it consistently for years, only delaying on a few special occasions for reasons like not having an internet connection, being in a different timezone, traveling, and some other situations that make a good enough to myself.
I'll keep pushing, and as my initial intention with this project was, I'll try to schedule more than one post per week to give myself a bit of slack to develop ideas more deeply and put more thoughts before I hit send.
Still, this project is for me to explore, and I'll continue to publish even if ideas aren't complete. There's always the following week to correct or expand on it.
See you soon!
You can join the newsletter here.
Researchers develop algorithms — the so-called neural networks and machine learning models.
Software engineers train and deploy those models and make interfaces for them.
Users interact with the artificial intelligence through machine-learning-powered applications.
Last week, I tested Luma for a podcast release. Luma is a platform for people to register for your events (for free or with paid tickets). Although I started using the platform to schedule my live streams a few months ago, this was the first time I used it for the podcast.
I was gathering interest for an incoming podcast release that would take place on YouTube in the form of a scheduled premiere, as YouTube calls uploaded videos scheduled to stream for the first time. During the video premiere, your audience can participate in the chat as the video plays, which, in our case, allowed listeners to ask questions to our latest guest, Andy Payne. After the video "premieres," it will be available on YouTube like any other on-demand video you upload. The gist is that, during the premiere, everyone watches the video simultaneously as if it were a live event.
Luma will remind your attendees when the event occurs in their timezone one day before and one hour before, sending them the link and instructions you set up.
Upon registration to an event, attendees have to enter their name and email, which allows you to invite them to similar events in the future or to reach out to them directly, for instance, when new content related to the episode they signed up for is available, i.e., the audio version of the podcast on Spotify and Apple Podcasts.
You can change the location link of the event so that attendees will receive it when the event is about to happen. If you change your mind before the event, you can direct users to a different platform (say, Twitch or Riverside). If that were to happen and you've directed users to YouTube, you can do nothing.
I shared the Luma link on a LinkedIn post, a tweet, and this newsletter and got twenty-two sign-ups from people I don't know, which isn't bad.
People can subscribe to you on Luma to get notified about future events you schedule. (Here's my Luma profile.) It's a great alternative to newsletters to capture interest in specific events. People may not be willing to sign up for your newsletter, but they might for the events they want to attend.
I met Andy Payne in the summer of 2016. Autodesk had acquired Monolith1 (a voxel-based editor) from Andy and Pan earlier that year, and I joined them as an intern to build a generator of 3D-printed material gradients and play with a Zmorph 3D printer.
I interned at Autodesk’s Boston Seaport office with Jose Luis García del Castillo and Keith Alfaro. Jose Luis and I would bike to Central Square, ride the Red Line T to South Station, and hop onto the Silver Line Bus to Drydock Avenue. By the end of the summer, I open-sourced Voxel2GCode2—a set of workflows to transform geometric objects and voxel-based models into 3D-printable G-code instructions.
Andy is an architect and software developer at McNeel, the company behind Rhino and Grasshopper 3D. We recorded a podcast conversation in New Orleans in September 2022, where I learned about Andy's new adventure.
Enjoy this episode on the origins of Grasshopper, Grasshopper 2, Rhino.Compute, teaching, learning to code, generative AI, open-source code and monetization, and Andy's journey.
We'll premiere the episode on YouTube on Wednesday, March 27, at 2 PM ET, which means you'll be able to chat with Andy and me during the episode.
You can RSVP here.
Monolith is a voxel-based editor that sits somewhere in between 3D CAD apps like AutoCAD and Rhino and pixel-based image-editing apps such as Photoshop. One of Monolith’s strengths is its capability to produce multi-material 3D prints with varying material densities. ↩
You can find Voxel2GCode on GitHub. ↩
I export videos from Descript, which has embedded subtitles, and Descript doesn't have a way to export subtitles by chapter markers; it only exports them for an entire composition.
Here's a command that extracts the embedded subtitles from a given video—and supports any format supported by FFmpeg, such as MP4, MOV, or MKV.
ffmpeg -i video.mp4 -map 0:s:0 subtitles.srt
Here's what each part of the command does.
-i video.mp4 - the input file.map 0:s:0: - maps the first subtitle track found in the video. (You can change the last digit to extract a different track, e.g., 0:s:1 for the second subtitle track.)subtitles.srt - the output file name and format, e.g, SRT or VTT.If you found this useful, let me know!
In Live 116, I conducted a live work session learning how to fine-tune Stable Diffusion models with Low-Rank Adaptation (LoRA).
If this interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next time!
01:07 · Introduction
01:21 · Today
02:19 · Fine-Tune with LoRA
04:09 · Image Diffusion Slides
06:43 · Fine-Tune with Lora
13:31 · Stable Diffusion & DALL-E
22:27 · Fine-Tuning with Lora
01:34:20 · Outro
I check whether my Tuesday post is ready.
As of lately, the answer is usually, No, the post is not ready. I then browse my archives or write something completely from scratch.
Why? I decided to do so, and I've experienced how it pays off in the long run.
It's craft, not art.
"A pro views her work as craft, not art." Says Steven Pressfield in The War of Art.
It's professional, not amateur.
"Professionals stick to the schedule; amateurs let life get in the way." Says James Clear in Atomic Habits.
Good Mondays, I enjoy the freedom of having completed my "homework" in advance.
Hi! It's Nono. Here are links to things I mentioned in my guest lecture at the Creative Machine Learning Innovation Lab at Berkeley MDes, invited by Kyle Steinfeld on March 15, 2024.
In Live 115, we played with tldraw's 'Draw Fast' experiment, which turns freehand scribbles and shapes into realistic images using the Optimized Latent Consistency (Stable Diffusion v1.5) machine learning model through fal.ai's API.
Thanks to the tldraw team for open-sourcing this experiment. ❤️
If this interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next time!
00:17 · Introduction
02:30 · Today
04:17 · Draw Fast by tldraw
06:15 · Fal AI
07:20 · Hands-On Draw Fast
08:03 · What is Draw Fast?
10:09 · Clone Draw Fast
14:16 · Fal AI
15:04 · Sign Up
16:41 · API Key
20:17 · Pricing
21:55 · DEMO
25:55 · Credits
28:03 · Models
30:57 · DEMO
37:59 · Challenge
41:27 · Break
44:42 · Tldraw React component
49:23 · Draw Fast Code
01:05:50 · Outro
I was sad to see a redirect from Lobe.ai to Lobe's GitHub repository.
Thank you for the support of Lobe! The team has loved seeing what the community built with our application and appreciated your interest and feedback. We wanted to share with you that the Lobe desktop application is no longer under development.
The Lobe team open-sourced a lot of their tooling to use Lobe-trained models on the web, or with Python, .NET, and other platforms. Yet the Lobe app and website were never open-sourced, which means they will no longer be usable when they cease to work.
Before it's gone, you can access Lobe's site and download the latest app at aka.ms/DownloadLobe.
Lobe takes a new humane approach to machine learning by putting your images in the foreground and receding to the background, serving as the main bridge between your ideas and your machine learning model.
Lobe also simplifies the process of machine learning into three easy steps. Collect and label your images. Train and understand your results. Then play with your model and improve it.
I'd invite you to listen to my conversation with Adam Menges on the origins of Lobe.
We often say, "I have no time," instead of "It's not a priority."
In 168 Hours, Laura Vanderkam states that, in our society, "there is plenty of time for raising kids while running a business, for working, teaching, and training for a triathlon, or whatever brings joy and meaning to your life."
Vanderkam says that recognizing this requires changing the narrative: instead of saying we don't have time to do X, Y, or Z we should say we won't do X, Y, or Z because "it's not a priority."
In its policy update of February 21, 2024, PayPal announced that it will exclude NFTs from eligibility for the buyer protection program and limit the protection of sold NFTs to ten thousand dollars at the moment of the transaction.
We are revising PayPal’s Buyer Protection Program to exclude Non-Fungible Tokens (NFTs) from eligibility [and the] Seller Protection Program to exclude from eligibility Non-Fungible Tokens (NFTs) with a transaction amount of $10,000.01 USD or above (or equivalent value in local currency as calculated at the time of the transaction); $10,000.00 USD or below (or equivalent value in local currency as calculated at the time of the transaction), unless the buyer claims it was an Unauthorised Transaction and the transaction meets all other eligibility requirements.
The crypto world seems to be the perfect place for fraudulent and counterfeit transactions, as scammers request money through digital wallets, which are often hard to trace and have no protection, instead of using traditional banks.
This policy update comes right after Cent NFT blocked NFT sales and the UK authorities seized, for the first time, three NFTs.
Every endeavor has a before and an after.
Before and after it's done; before and after you give up; before and after you have energy for it; before and after you have time for it.
I try to do whatever I must get done in the day before I run out of time and energy.
Not everything can be done in one go, even if you set yourself to it, and it's easy to fall into the trap of postponing for later.
Doing things you know you'll end up doing today as early as possible always works best.
In the morning, your mind is sharp, and you're energized.
At night, as happens to me many days with writing, each word requires ten times more effort than when I woke up.
It's a matter of priorities and focus.
One of the ideas I wanted to build in 2024 was a daily word counter to track whether I had written my two hundred words of the day.
I made a Mac menu bar app1 that displays the word count below the running date—say, 240227, for today—and shows the ✍🏻 emoji until I’ve passed the two-hundred-word mark.
As I mentioned last week, I write daily within a fuzzy timeframe. As the app doesn’t know when I wake up or go to bed, I programmed it for the day to start and end at 6 am.
If you’re curious about why I write daily and how I do it, you can learn more about my writing habits.
How to run Google Gemma 2B- and 7B-parameter instruct models locally on the CPU and the GPU on Apple Silicon Macs.
In Live 113, we ran Google's Gemma LLM 2B- and 7B-parameter open models on an Apple Silicon Mac, both on the CPU and the GPU.
We downloaded the Instruct models with the Hugging Face CLI and used PyTorch with Hugging Face's Transformers and Accelerate Python packages to run Gemma locally.
If this is something that interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next time!
01:23 · Introduction
02:46 · Previously
03:11 · Today
03:45 · Elgato Prompter
06:19 · Interlude
06:43 · Google Gemma 2B & 7B
08:45 · Overview
11:59 · Hugging Face CLI
14:01 · CLI Install
14:54 · CLI Login
15:33 · Download Gemma
22:19 · Run Gemma Locally
24:49 · Anaconda Environment
29:00 · Gemma on the CPU
52:56 · Apple Silicon GPUs
55:32 · List Torch Silicon MPS Device
56:50 · Gemma on Apple Silicon GPUs
01:08:16 · Sync Samples to Git
01:17:22 · Thumbnail
01:28:42 · Links
01:31:12 · Chapters
01:36:28 · Outro
Performance Max campaigns serve across all of Google’s ad inventory, unlocking more opportunities for you to connect with customers.
[…]
[Google announced] several new features to help you scale and build high-quality assets — including bringing Gemini models into Performance Max.
[…]
Better Ad Strength and more ways to help you create engaging assets.
[A]dvertisers that use asset generation when creating a Performance Max campaign are 63% more likely to publish a campaign with Good or Excellent Ad Strength.
For digital systems, the day starts and ends at twelve o'clock at night—every day is twenty-four hours (00:00–24:00).
For humans, the day begins when you wake up and lasts til you go back to sleep.
Days at home are commonly sixteen hours—from eight in the morning til midnight.
Traveling days are shortened or extended depending on whether you travel east or west, lasting up to thirty hours when you fly from Europe to California.
It is within this fuzzy timeframe that I write and sketch daily.
In Live 112, we did a hands-on example of how to deploy a web app with Vercel.
We used Yarn Modern (4.1.0) to create, develop, and build a Vite app that uses React, SWC & TypeScript, pushed the app to GitHub, and import the Git repository into a Vercel deployment, which then re-builds and deploys on every code change.
If this is something that interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next time!
00:16 · Introduction
01:58 · Previously
02:26 · Today
05:21 · Diffusion Models for Visual Computing
10:07 · LGM
11:21 · Interlude
12:53 · Vite, React & TypeScript Apps with Yarn Modern
17:20 · Create the App
24:29 · Push to Git
29:07 · Deploy to Vercel
33:40 · Edit the App
42:53 · YouTube Channel
45:23 · Draw Fast
46:25 · Markers
47:51 · Elgato Prompter
48:27 · Markers
51:45 · Outro
I kept seeing this error when creating a new Yarn Modern project—setting Yarn to the latest version with yarn set version stable—and using the yarn init command.
Usage Error: The nearest package directory doesn't seem part of the project declared in […]
For instance, yarn add -D @types/node wouldn't work.
There was a package.json and a yarn.lock file in my home directory.
Removing the file fixed the issue.
rm ~/package.json ~/yarn.lock
Then, in any directory, even subfolders of your home directory (~) you can create new yarn projects.
mkdir app && cd app
yarn init -y
yarn add -D @types/node
When you run yarn set version stable, Yarn Modern creates a package.json with the packageManager property set to the latest stable version of Yarn, such as 4.1.0.
To avoid the above issue, you should first create your project.
yarn create vite my-app --template react-swc-ts
Then, enter the app's directory and only then set the desired Yarn version.
cd my-app
yarn set version stable
yarn
# Yarn Modern will be used here.
...is easier than it sounds.
But also harder to accomplish consistently if you overthink it.
I keep writing and journaling daily.
Yet I've experienced how stopping a habit for a few days can turn into stopping it for a very long time—weeks, months, years.
The gist is to put systems in place that make you do what you should do, to spend time doing impactful things.
After I put my M1 MacBook Pro to sleep for a couple of weeks, it woke up with the wrong date and time, set to around two years before the current date.
Here's what fixed it for me.
Open a Terminal window and run the following command.
sudo sntp -sS time.apple.com
This will trigger a time sync with the actual date and time from Apple's clock. But it won't completely fix the issue. If you close the lid and reopen your laptop, the time will return to the wrong time.
After you run the command above, you have to do the following.
That's it. This permanently fixed the issue for me.
If you found this useful, let me know!
You can directly assign new properties.
(window as any).talk = () => { console.log(`Hello!`) }
You can extend the Window interface with typings and then assign the property values.
declare global {
interface Window {
talk: () => void
concat: (words: string[]) => string
}
}
window.talk = () => { console.log(`Hello!`) }
window.concat = (words: string[]) => {
return words.join(`, `)
}
I don't buy the promise behind 1Password or LastPass.
You only need to remember one password.
The last password you'd need to remember.
They don't tell you that you're also building a one-stop shop for hackers to steal it all at once.
The solution?
Store hints, not passwords.
Don't reuse passwords. Use algorithmic passwords instead.
Use passkeys and security keys.
In Live 111, I showed a few tools I've recently discovered.
If this is something that interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next time!
00:11 · Introduction
02:34 · Previously
03:54 · Password Managers
06:45 · Notion
07:57 · Animations with Lottielab
13:33 · Animations with Linearity Move
17:31 · Visual Electric: Generative AI Images
21:32 · Break
23:25 · Visual Electric
26:27 · Future Topics
27:03 · Outro
Squirrels scatter collected foods in numerous locations to minimize the risk of losing everything at once. Certain animals even collect and cache until food ripens and becomes edible.1
Everything under the same roof can be stolen with one key.
Centralization is convenient but unsafe.
Decentralization takes more work but is more secure.
It's all about minimizing risk.
Hoarding (Animal Behavior). Wikipedia. Accessed Jan 27, 2024. ↩
In an email to Apple Podcasts Connect, Apple announced today they will be displaying transcripts of podcasts in Apple Podcasts, which will be generated by them.
To make podcasts accessible to more users, we’re adding transcripts to Apple Podcasts. Listeners will be able to read along while your podcast plays or access a transcript from your episode page.
Apple Podcasts will start to automatically create and display transcripts for your shows, or you can provide your own. Learn more about our newest feature.
This is a great feature which, if their transcripts are accurate enough, may render useless the countless hours of manual transcription or manual AI-transcript editing to make sure the produced transcripts are accurate.
I wondered whether Apple would allow podcast creator edits when AI gets parts of the transcripts wrong. The answer is yes; transcripts can be provided via podcasts' RSS feeds or uploaded to Apple Podcasts.
Apple will automatically generate episode transcripts for your show. You can also provide your own transcripts through RSS or upload. Displaying transcripts will make your podcast more accessible.
Apple lists two options. "Only display auto-generated transcripts by Apple" and "Display transcripts I provide, or auto-generated transcripts by Apple if one isn't provided."
If you choose to provide your own transcripts, we will ingest them using the RSS transcript tag. […] All transcripts are subject to quality standards. Files that do not meet standards will not be displayed.
You can download and edit transcripts in Apple Podcasts Connect, make changes, and link the new file to your RSS feed.
For this, Apple has added the new tag <podcast:transcript> to specify "a link to the episode transcript in the Closed Caption format. Apple Podcasts will prefer VTT format over SRT format if multiple instances are included."
You should use this tag when you have a valid transcript file available for users to read. Specify the link to your transcript in the url attribute of the tag. A valid type attribute is also requried. Learn more about
namespace RSS tags on the Github repository. Options for displaying transcripts are available in Apple Podcasts Connect for each show.
I'm looking forward to this amazing feature. It will make the vast catalog of existing podcasts more accessible.
In Live 110, I continued looking at Apple's MLX framework.
Watch this stream to learn how to run MLX code in Python and generate text with Mistral 7B on Apple silicon.
If this is something that interests you, please let me know on Twitter or, even better, on the Discord community.
Thanks for watching.
See you next week!
00:17 · Introduction
02:35 · Today
04:35 · Apple MLX
06:40 · mlx
08:24 · mlx-data
09:55 · mlx-examples
10:43 · MLX Community in HuggingFace
13:40 · M1 Pro with MLX?
15:43 · mlx-lm Troubleshoot
26:19 · mlx-lm Solution
31:57 · Lazy Evaluation
34:09 · Indexing Arrays
39:48 · Generative Image Control
40:48 · Instruct Pix2Pix
45:21 · ControlNet Depth
52:47 · LLMs in MLX with Mistral 7B
A faraway deadline gives you time to think, work, and procrastinate.
No deadline makes you feel you have all the time in the world, which you don't.
A short deadline induces you to focus and makes you more efficient.
And a super-short deadline adds pressure and freezes you.
The sweet spot is to fake short deadlines to ship often.
That way, you can't hide, procrastinate, and won't freeze because you'll be ready when the time comes.
